Method validation - accuracy spiking approach

[Robotjock]

Virtual colleagues

I kindly request your input/comments/best practices regarding a discussion/debate occurring at my workplace. The issue concerns how to spike solvents into an API for accuracy validation.

Assuming the API contains some residual solvents (as they usually do), do you….

1) Spike TO a level, i.e., 100 % of spec. That is, add the appropriate amount of solvent plus what the API contains to equal 100 %.

The rationale for this approach is that if you spike AT the level (described below) and include any residual present the combined response is greater the 100 % (or whatever level you’re testing). Thus, it is not a “true” test of accuracy.

2) Spike AT the level. That is add the appropriate amount of solvent to equal 100 %, then correct (subtract) for any residual present in the API.

The rationale for this approach is, 1) you don't accurately know beforehand the amount of residual present (no validated method) to correctly spike to 100 %, 2) a blank correction is common practice and 3) the linearity validation data would substantiate any combined response

Assumption – the residual solvent is not present in large amounts, i.e., you would not be subtracting a large background from a small signal.

Thanks in advance.
Robotjock

[chromatographer1]

The presumptions here are unsettling. I would not agree to their soundness.

You first need to prepare a matrix that contains none of the solvents to truly perform accuracy tests.

Then you can spike at a known level. Unless your solvent impurity has extremely low volatility this should be possible with ease. If it has too low a volatility, there should be means to separate the matrix in small amounts from residual solvents usually found in manufacturing.

But using only one point is advisable? Would you not want a test at the minimum level and at the middle of the range? If your answer is no, then I ask (as would a regulatory reviewer) Why not?

I have performed Accuracy determinations by looking at a linear progression of several levels below through at least twice the proposed range of measurement and comparing that line to measurements of the matrix with known amounts spiked into a zero level sample.

Your test proposals have too many assumptions and inaccuracies for me to sleep well at night.

If all this is unreasonable or impossible in your lab, then you need to spike a singular sample of matrix over a series of levels below - through - above your measurement range (several portions of a single dissolved sample) and subtract the calculated determined amount of solvent present before spiking with your standard additions to prepare your linear regression line. Even if your sample has 100% of the proposed solvent range the highest level will be twice the range limit. And that level should be with the linear range of the method. (your claims should be conservative, not stretching the envelope of the method)

Your validation should have all these points in the range of your method. Once you know the residual amount of solvent present then you can compare whatever spiked sample (with determined and known amount of solvent apiked) to the regression line and determine the difference between the sample and the line as the error of measurement, or accuracy.

But good luck with the path you choose to take.

Rod